Replit drops Anthropic models to focus on OpenAI

AI development platform Replit is severing ties with Anthropic and shifting its focus toward OpenAI models, according to comments made by chief executive Amjad Masad at TechCrunch Disrupt. Masad cited a mix of pricing, collaboration dynamics, and benchmark results, noting that ChatGPT now yields better performance. Replit's internal evaluation tools alongside its public benchmarks demonstrated that models such as Sol and Astra delivered superior results when generating vibe-coded applications from scratch.

Although Replit formalized a partnership with OpenAI in August, the transition away from Anthropic remains somewhat disputed. Insiders familiar with the arrangement contend that the relationship is still active. Furthermore, recent company documentation and social posts listed Anthropic systems—specifically Claude Fable 5, Claude Fable 5.1, and Opus 5.5—as active components within the Replit Agent production workflow.

The shifts in Replit's underlying infrastructure coincide with rapid corporate growth. While the firm has not disclosed its annual recurring revenue, Masad stated that enterprise clients now account for approximately 15% of revenue, up from just 2% last year. Business-to-business sales as a whole now generate more than 50% of Replit's total income, demonstrating a clear pivot toward commercial contracts.

For software engineering teams, this model migration highlights how rapid shifts in LLM performance benchmarks dictate vendor choices in developer tooling. Masad also stressed the importance of creating a stable environment for autonomous coding assistants, arguing that AI agents require a structured "habitat" equipped with strong sandboxing, orchestration, error recovery, and strict identity authorization before being deployed safely into enterprise software workflows.